[kbaggen]

Hi,

I need a tester, does not matter, which Baytrail System. We are currently thinking of giving VPP, the vaapi postprocessing lib a second chance. sraue of OpenELEC has put together a build, which you can use for testing. The advantage of VPP is only used when you watch LiveTV which is interlaced.

VPP implements three different methods for Deinterlacing:
- Simple Bob (IVB, HSW)
- MADI - Motion Adaptive Deinterlacing (IVB, HSW)
- MACI - Motion Compensation Deinterlacing (HSW)

As the Baytrail is much too slow to copy the decoded surfaces back to system memory and then do some advanced deinterlacing, this VPP is highly useful.

You can download the image here:
http://snapshots.openelec.tv/test/OpenEL...063daa.tar

Testing Samples:
Transformers (576i): https://dl.dropboxusercontent.com/u/5572...formers.ts
Servus TV HD (1080i50): https://dl.dropboxusercontent.com/u/5572...g-test.mp4

How to test:
Download sample Transformers and Servus TV HD, play it, navigate to the film reel icon and select: Deinterlace Auto -> Deinterlacing Method: VAAPI (BOB), Scaling: Bilinear.
Press "o" and see if you get 50fps, if there are skips and drops (count only the numbers that appear from that point). Watch sample till the end
Now watch the sample again and select Deinterlacing Method: VAAPI (MADI)
Press "o" and see if you get 50fps, if there are skips and drops (count only the numbers that appear from that point). Watch sample till the end

This was the test for the 576i Sample.

Now repeat the above test with the 1080i50 sample.

Thanks very much
Peter

On Haswell: All methodes despite BOB are broken - which was the reason we abondonnned it long ago

Did some testing on the Q1900-ITX.
I couldn't find the VAAPI (MADI) option. There were only VAAPI BOB, VAAPI AUTO, BOB and BOB (Inverted) deinterlancing methods.
The VAAPI BOB and VAAPI AUTO had same results.

VAAPI BOB:
576i sample: Frame rate remains smooth @ 50 FPS, no noticeable drops or skips. Average CPU usage throughout 4 cores is about 10%. Most of the load goes to Core #1 and Core #2, Core 3 & 4's barely get any load (2-5% avg)
1080i sample: Frame rate jumps all over the place, from 5 to 60 and so on. Constant dropping and skipping. Video plays for 1 second then freezes for about 2-3 seconds and so on. Average CPU usage throughout 4 cores is about 15% with the #1 core doing most of the work (~20%ish)

Couldn't test with the VAAPI MADI since I did not find the deinterlancing method.

Those are really, really bad results :-(
That means that the GPU is too slow to do Deinterlacing and Decoding - ouh man - I would never have expected that.

Can you post an xbmc.log please from bootup and with those samples? Are you sure that scaling was set to Bilinear?

Background: We are currently using swfilter in OpenELEC. This copies the surfaces back from the GPU to system memory and there we deinterlace on the CPU. With that method that Baytrail Chip can do 1080i50 with BOB quite fine.

That's a desaster :-(

fritsch,
I ran those 2 samples on my "NUC altenative inspired" system yesterday J1900, eg. win7, hence I assume you cannot use it for much, but with the following specs I got 50 fps, 1 or 2 skips if I recall!.

*Deinterlace (OSD video setting), DVXA BEST or BOB (same fps)
*Upscalering (OSD video setting): DVXA
*Under System--> Video -->
1. Hardwareaccelation ON --> Pixel-shardes
2. 10%, upscalere .
3. Hardwareaccelation kan use DVXA2

(Also adjusted the framerat to movies = ON, at start/stop)

I also used LAcross3 (only posible for the i576 sample), and it gave also 50fps........If I understood your former topic that should not be poosible?

In all it looked very smooth!